mello-retrieval-accuracy-degrades-with-k
IN premise — summaries/2026/08/24/zhong-2023-mquake-sR-references-chunk-2.md
Created 2026-08-25T02:59:07+00:00
MeLLo's retrieval accuracy drops from 88.1% at k=1 to 59.7% at k=3000, directly reducing multi-hop accuracy from 52.1% to 30.9%.
Summary
The more documents MeLLo pulls in to answer a question, the less reliably it finds the right ones, and this directly weakens its ability to chain together multi-step answers. In practical terms, scaling up retrieval context doesn't help the system reason over longer question chains — it makes them significantly less likely to be answered correctly.